Design and development of low-power, long-range data acquisition system for beehives - BeeDAS

Abstract

Decision making capability of a system is highly dependent upon the quality and quantity of training data. Majority of beehive monitoring systems developed for research purposes are designed to collect data through a small set of sensors, and from locations with little geographic diversity. This hinders the development of a dataset that can be used to effectively train machine learning models. In this work, we explain the design and development of a multi-sensory, remote data acquisition system for beehives (BeeDAS), with focus on low-power consumption and long-range communication. We address design challenges associated with such systems and highlight the critical issues that need consideration. The proposed system enables collection of data from beehives at remote locations and harsh environment. Results of field deployments elucidate the effectiveness of various sensors which measure temperature, humidity, atmospheric pressure, CO2, acoustics, vibrations and the weight of a hive in hostile environment. This work also uses random forest regression to evaluate the feature importance of different sensors, environmental variables such as temperature, humidity, rain, wind speed as well as the information related to seasons, towards estimating the daily hive weight change, on a dataset comprised of 1,250 days of sensor recordings. We also evaluate the protocol designed for communication using Narrow Band Internet of Things (NB-IoT). The issues related to power optimization, sleep intervals and data storage in remote monitoring are also discussed.